Zeromatter has emerged from stealth after four years of development on a platform for training and testing physical AI, according to November 12, 2026 posts by omarsar0 and rohanpaul_ai. They describe a shared simulated world where cars, drones, aircraft and robots can practice before operating in the real world. Let the virtual world have the terrible day. Source Source
rohanpaul_ai says Zeromatter built its own renderer, physics engine and simulation platform around how cameras, lidar and radar perceive their surroundings. The same post says a given test produces the same result every time, and that simulation results are checked against real data from the company’s own labs. Source
rohanpaul_ai says rare scenarios can be rerun on demand. If the platform works as described, teams could test changes against the same troublesome conditions. Waiting for a rare failure is a terrible test schedule. Source
But a repeatable test isn’t automatically a realistic one. Neither post supplies technical documentation, benchmark results or an independent evaluation establishing the platform’s physical accuracy or how well training in simulation carries over to real machines. Checking results against the company’s own lab data, as rohanpaul_ai describes, is not the same as independent validation. Source Source
Neither post establishes pricing, access terms, customers or general availability. Emerging from stealth doesn’t mean anyone can sign up and start rehearsing robot mishaps. Source Source
Comments